---
title: Use Ollama with Tactigon SHAPE
description: Connect Tactigon SHAPE to Ollama and use visual blocks to send prompts and receive responses from a locally running AI model.
url: "https://thetactigon.com/docs/use-ollama-with-tactigon-shape/"
updated: "2026-11-05"
category: AI INTEGRATION
---

# Use Ollama with Tactigon SHAPE

Use Ollama with Tactigon SHAPE to send prompts to a locally running language model and display its responses through visual programming blocks.

In this guide, you will configure the connection, build a prompt workflow and check the generated response in SHAPE’s output terminal.

## How the integration works

The example uses three components:

- **Tactigon SHAPE** connects visual blocks to send a prompt and display the response.
- **Ollama** runs the language model and processes requests from SHAPE.
- **Mistral** generates the text response.

Mistral is the model used in the supplied example. The model name configured in SHAPE must match a model available in your Ollama instance.

## Before you begin

Make sure that:

- Tactigon SHAPE is installed and running. If needed, follow [Install SHAPE with Docker](https://thetactigon.com/docs/install-shape-docker/).
- Ollama is running and reachable from the SHAPE environment.
- The Mistral model is available in your Ollama instance.
- You know the Ollama service URL and the exact model name.

Use an Ollama address reachable from the environment running SHAPE. If SHAPE runs inside a container, do not assume that `localhost` refers to the computer hosting Ollama.

## 1. Create and configure your Shape

Click **Create Shape** and complete the following fields:

| Field | What to enter |
| --- | --- |
| **Name** | A name for your project, such as `Ollama first prompt`. |
| **Description** | An optional description of the workflow. |
| **Ollama — URL** | The address of your Ollama service, reachable from SHAPE. |
| **Ollama — Model name** | The exact installed model name. Enter `mistral` if this matches your Ollama configuration. |

Click **Create Shape** to open the workspace.

![Create Shape dialog in Tactigon SHAPE with Ollama URL and model name fields.](https://thetactigon.com/wp-content/uploads/2026/10/ollama-shape-configuration-765x1024.jpg)
Enter the Ollama service URL and an available model name when creating your Shape.

## 2. Build the prompt workflow

Here's how to connect an **AI prompt** block to **Debug**, inside a **Loop** block.

1. Place a **Debug** block inside the **Loop** block.
2. Connect an **AI prompt** block to the input of **Debug**.
3. Connect a text block to **AI prompt**.
4. Enter the question or instruction you want to send to Mistral.

The blocks have these roles:

| Block | Role |
| --- | --- |
| **Loop** | Contains the workflow that runs repeatedly. |
| **AI prompt** | Sends the connected text to the configured model and returns its response. |
| **Debug** | Displays the response in the output terminal. |

If you take a look at the next screenshot you can see question about sensors and the generated answer. It demonstrates the connection between the prompt block and the displayed output.

For a short first test, you can replace the text with this prompt:

```
Explain MQTT to a beginner in exactly three short bullet points.
Cover the broker, publishers and subscribers.
Reply in English.
```

This alternative prompt uses the same block connections as the illustrated example.

![AI prompt and Debug blocks inside a Loop in Tactigon SHAPE, with the generated response in the output terminal.](https://thetactigon.com/wp-content/uploads/2026/10/ollama-shape-prompt-response-1024x196.png)
Connect AI prompt to Debug to display the model response in the output terminal.

## 3. Save and run

Save your Shape and start it using the execution control.

If you need help creating, saving or running a workflow, see [Create your first Shape](https://thetactigon.com/docs/create-your-first-shape/).

Watch the output terminal for the response displayed by **Debug**.

## 4. Check the result and stop the workflow

A successful first test should show:

- a generated text response in the output terminal;
- content related to the prompt you entered;
- no connection or model errors preventing the response.

If you use the MQTT test prompt, check whether the response explains the broker, publishers and subscribers.

The wording may vary between runs, but a response that does not follow the requested format exactly does not necessarily indicate a configuration problem.

Because the example uses **Loop**, the prompt may be sent repeatedly. Stop the Shape using its execution control after checking the first response.

## Troubleshooting

| Issue | What to check |
| --- | --- |
| **No response appears** | Confirm that Ollama is running and that its URL is reachable from the SHAPE environment. |
| **The model is unavailable** | Check that the model name in SHAPE matches a model installed in Ollama, including any version tag. |
| **The response is not displayed** | Check that **AI prompt** is connected to **Debug** and that the workflow is running. |
| **Responses keep appearing** | The request is inside **Loop**. Stop the Shape after completing the test. |
| **The answer differs from the example** | Generated responses can vary. Check relevance to your prompt rather than expecting identical wording. |

For an installation managed through Docker Compose, open a terminal in the directory containing your Compose file and inspect the services:

```
docker compose ps
```

To view their logs:

```
docker compose logs -f
```

Look for connection errors, unavailable models or service startup failures. Press **Ctrl+C** to stop following the logs; this does not stop the Shape workflow.

## Next steps

Try changing the prompt while keeping the same block connections. Run the Shape again and check how the generated response changes.

For the source instructions and screenshots, see the [Tactigon SHAPE Ollama guide](https://github.com/TactigonTeam/Tactigon-Shape/wiki/Ollama?utm_source=chatgpt.com).